Field of the Invention
The aspect of the embodiments relates to character recognition processing.
Description of the Related Art
Conventionally, in recognition processing to be executed on characters on paper, an outline of a character is detected, a position of a character image is identified, and a character row in a horizontal or vertical direction is recognized. Then, an inclination of the recognized character row is corrected, a character image of each character is cut out from the character row, and each character is identified from the cutout character image. Therefore, if a character cutout position is inappropriate, a correct character cannot be recognized.
Japanese Patent Application Laid-Open No. 2004-46723 discusses a technique for obtaining a character arrangement pitch of cutout characters, and detecting, as a character cutout error, a portion where an irregular character arrangement pitch greatly deviating from an average value is caused.
Further, in recent years, wide usage of smart phones and digital cameras has enabled image information including character information to be easily acquired. As a result, there is a growing market for capturing characters from a wide variety of measurement environments. For example, there is such a use case that, in a quarry of mines, serial numbers engraved on tires are used for managing the tires used on dump trucks. In such a case, it is considered that the serial numbers engraved on the tires are captured by smart phones or digital cameras, the captured images are then subjected to character recognition processing, and the tires are managed by using the serial numbers as character recognition results. However, in the case of serial numbers engraved on the tires, contrast between characters and a background may be low or surfaces may be much soiled, and thus preconditions for such a case differ greatly from those for a case where characters on paper are recognized.
When the contrast between characters and a background is low or when the surfaces are much soiled, it is difficult to accurately detect character rows based on outlines of the characters because detection accuracy of the outlines is low. In addition, in images captured by smart phones, imaged subjects are more likely to be inclined than a case where documents are scanned by flatbed scanners. When images with inclined subjects are subjected to character recognition without considering the inclination, recognition accuracy is deteriorated.